Verdict · fail
Funding capture did not survive 2026.
We wrote the pass/fail rule on 21 Sep 2026, before running the test. The rerun on 25 Sep 2026 failed it. The research harness is frozen; every number stays published.
The question
Over the full backtest, a delta-neutral position (long spot, short perpetual) collected more funding than it paid. After trading costs the edge was thin. The question for this test: does it survive realistic costs and a margin test in the most recent market, or only on a four-year average?
The rule, written first
The thesis advances only if at least one non-zero-cost cadence (weekly, biweekly or monthly) is positive in the first half of 2026 and survives the default margin test, on complete days only, at 5 bps fee per side + 5 bps slippage. Otherwise the harness is frozen.
Written 2026-09-21 · run 2026-09-25 · result: fail
The result
Only monthly rebalancing at a lower 2 + 2 bps cost came out positive (+0.01%). That is below the cost the rule names, so it does not count.
The margin test
A hedged pair can still be closed out: the short perpetual leg is margined on its own. The test opens a fresh account at the start of each window at 5× (20% initial margin, 5% maintenance) and applies one 2% basis shock on the worst day. Over the whole 1 Jan 2022 – 3 Sep 2026 window, none of 16 cost and rebalancing combinations survived it from a 2022 start.
What this does not say
- It is a historical backtest, not a forecast and not a recommendation to trade or not to trade.
- Sub-periods are slices of one run; nothing was re-fitted to them.
- Funding is summed per UTC day, not settlement by settlement.
- Exchange, counterparty, capacity and financing costs are not modelled.
- Whether low funding in 2026 is a cycle or a lasting change is not known. No cause is claimed.
What would reopen it
The same rule on a later window: a weekly, biweekly or monthly result that is positive after 5 bps fee per side + 5 bps slippage and survives the same margin test, on complete days only. Nobody is running it now. If it is ever rerun, the result goes on this page whether it passes or fails.
Why publish a test that failed
A results page that shows only what worked is a selection. This is the other half: the rule, the run and the miss. The full tables are on results, the strategy rejected earlier is on compare, and the questions to ask of any backtest are in the backtest checklist.
Backtest and simulation only — not financial advice.